IEEE Consumer Electronics Magazine - January/February 2024 - 51

FIGURE 2. Risk analysis for some major types of cryptocurrencies.
manually and structural features can be identified
by deep learning techniques.
Feature Selection
This step is significant in data dimensionality
reduction by selecting a group of features that represent
accurately all extracted features. Many
techniques deal with this problemsuch as unsupervised
streaming feature selection (SFS) and twostage
feature selectionmethods. The SFS technique
manipulates a new feature according to its arrival
instead of searching for a set of all features, which
is a complex and costly process. For the two-stage
feature selection methods, the first stage includes
many filters depending onmethods of local feature
selection as well as several methods of feature set
construction, and in the second stage, different
methods are employed such as component analysis
(PCA), genetic algorithms and latent semantic
indexing.
Risk Control
Text mining is considered a well-known technique
for extracting information from several types
of data such as text, images, and videos on social
media websites. It includes many tools such as text
analysis, text processing, and knowledge discovery
to be aware of the response of people about specified
posts, as well as, the agree and disagree
opinions about specific topics. SNA involves determining
the social relationships between various
social media users based on the network theory of
nodes and connections. Sentiment analysis is a
type of NLP, text analysis, and statistics. The task is
to discover the sentiment of the text by categorizing
them into positive, negative, or unbiased. This
analysis is as a rule utilized for decision making,
i.e., users like or dislike something, or the item is
nice or awful. Modeling analysis of social media is
concerned with managing and assessing tools of
informatics for social media data gathering, examining
aswell as analyzing.
January/February 2024
CASE STUDY AND RESULTS
USA is one of the leading countries in an investment
of cryptocurrency and includes the highest
distribution of investors on important crypto
exchange and wallet Coinbase. According to some
reports, currently, 58% of Americans declare to
have heard of Bitcoin. Bitcoin has received more
Google searches in the past year than either the
royal wedding or election results. More than 70% of
US States have so far passed legislation addressing
cryptocurrencies or blockchain technology.
So, we focused on making the case study on
the American people who are interested in
cryptocurrency and the Twitter platform. We
collected more than 81k tweets based on the
geo-location of users through six months starting
from 01-02-2022 to 01-08-2022 by using the
Twint tool. The proposed model can be easily
employed for other social media platforms
because this model is based on NLP and can
deal with any text either on Twitter or on other
platforms. However, we focused on Twitter
especially, because it includes important
accounts for well-known users of cryptocurrencies
such as Elon Musk and Vitalik Buterin.
Hence, much data will be available and more
applicable.
To manage the risks in the dataset, first, we
will utilize risk analysis. It helps in examining
whether a tweet indicates no risk, potential risk,
or risk.
We centralize on analyze risks in three common
cryptocurrencies types namely, Bitcoin,
Ethereum, and Altcoin. We built a new dictionary
that includes more than 7500 words that refer to
a risk when it is mentioned in a text. For Bitcoin,
we analyze more than 59k tweets to identify risks
in each tweet. For Ethereum and Altcoin, we analyze
9.9k and 2k tweets, respectively. The results
are visualized in Figure 2. To validate the
obtained results, we use different performance
measures such as accuracy, precision, recall,
51
IEEE Consumer Electronics Magazine - January/February 2024

Table of Contents for the Digital Edition of IEEE Consumer Electronics Magazine - January/February 2024
